This thesis presents an in-depth exploration of advanced deep learning applications in particle physics, particularly in the context of tracking, calorimetry, and energy reconstruction within High Energy Physics (HEP) experiments. It encompasses three studies, each underscoring the potential of deep learning to address the increasing computational demands posed by more powerful particle accelerators and their complex datasets. The first study highlights the application of Graph Neural Networks (GNNs) in the Exa.TrkX project for efficient particle tracking. The second study focuses on the DeepCalo model, a multi-modal deep learning model incorporating FiLM layers, dense layers, and convolutional layers, adept at processing ECAL data, tracks, and high-level scalars. This study specifically demonstrates the novel implementation of DeepCalo on Field-Programmable Gate Arrays (FPGAs) for low-latency applications in particle physics experiments. The third study evaluates the use of Sparse Point-Voxel Convolutional Neural Networks (SPVCNN) for clustering energy deposits in hadronic showers, showcasing its potential for real-time data analysis in high-energy environments. Collectively, these studies not only exhibit the adaptability and computational efficiency of deep learning models in HEP but also indicate their critical role in managing the computational load of future high-luminosity experiments, which may be vital to addressing some of the most pressing challenges in modern particle physics, such as understanding dark matter.
Final Exam: Alex Schuy
